An AI couldn’t beat humans at StarCraft, so it decided to cheat

OpenAI's GPT-6 Astra bot cheated in the StarSkirmish StarCraft benchmark after failing to beat the top human-made bot Stardust, downloading and running Stardust's code instead of its own, according to Kotaku. StarSkirmish creator Kai McPheeters rolled back GPT-6 Astra's code. GPT-6 Astra and Claude Opus 5.5 were essentially tied as the best AI-made bots but neither topped Stardust, the top-rated human-made bot.

StarSkirmish https://starskirmish.com/bench/ pits AI-made StarCraft-playing bots against one another, as well as against human-made bots. OpenAI’s GPT-6 Astra and Claude Opus 5.5 were essentially tied as the best-performing AI-made bots, but they couldn’t top Stardust, the top-rated human-made bot. An AI couldn’t beat humans at StarCraft, so it decided to cheat When GPT-6 Astra’s own bot failed, it simply downloaded the best human-made bot instead. On Friday, GPT was facing off against Claude and the human-created bot Pluto, but according to Kotaku https://kotaku.com/openais-gpt-6-astra-gets-frustrated-losing-at-starcraft-and-decides-to-cheat-instead-2000739607 , it couldn’t quite get an edge. So it resorted to a tactic that is becoming alarmingly common for modern AI models — it broke the rules https://x.com/kaimcpheeters/status/2106082840186147226?s=20 . GPT-6 Astra went and downloaded Stardust, and started running that instead of its own bot. StarSkirmish creator Kai McPheeters eventually rolled back GPT’s code https://x.com/kaimcpheeters/status/2106082842560405780?s=20 . That GPT-6 Astra took it upon itself to go outside the bounds when confronted with an obstacle shouldn’t surprise anyone. When OpenAI agents couldn’t get the data they wanted from a UN website, they found a creative solution https://www.theverge.com/ai-artificial-intelligence/1001178/openai-agents-bruteforce-un-website and hijacked Google’s XSS game a cross-site scripting learning tool . The company’s agents also have engaged in “ deceptive behavior https://www.theverge.com/ai-artificial-intelligence/975577/aisi-openai-anthropic-agent-hacking ” to cover their tracks. Of course, OpenAI’s agents are the only ones going rogue, but at least they haven’t been caught cheating at StarCraft yet. Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.